How AI is shrinking engineering teams and rewriting tech role playbooks / AI正在重塑技术团队：从「两张披萨」到「一张披萨」的组织变革

For years, the tech industry has lived by Amazon's "two-pizza team" rule: if a team can't be fed by two large pizzas, it's too big. But a recent piece I came across argues we're entering a new era where even that size is overkill — the ideal engineering team now fits on a single pizza, at 2-3 people per project. That shift isn't random; it's a direct ripple effect of AI coding tools cutting down the time engineers spend writing, reading and debugging code, the historic bottleneck for most tech projects.

很长一段时间里，代码产出速度是决定项目进度的核心因素，但当 Claude、GitHub Copilot 这类工具把编码效率提升数倍后，我们突然发现卡脖子的环节变了：PM 的用户调研、需求梳理和设计师的原型产出，现在成了拖慢节奏的主要原因。更尴尬的是，传统团队配置里往往是 4-7 个工程师配 1 个甚至共享 1 个 PM 和设计师，这种资源不平衡在 AI 放大了工程师产能之后，被暴露得更加明显。

The most interesting response to this imbalance, to me, is the sudden rise of product engineer roles. This isn't a new title by any means — it's existed for over 16 years — but it's finally moving from a niche role to a mainstream need. Product engineers are engineers who step beyond pure coding to own parts of the product process: talking to users, drafting roadmaps, even putting together prototype interfaces using pre-built design system components. They don't replace PMs or designers, but they take over the execution-heavy parts of those roles, so the core product and design teams can focus on higher-impact, more creative work that AI still can't do well. It's like giving the team more hands to build with the Legos that designers and PMs put together.

但这并不意味着「全栈工程师」会变成主流，恰恰相反，行业对领域专家的需求会越来越高。AI 写的代码通常能达到合格线，但很少能做到优秀：它会不小心删掉关键逻辑，会复制代码库里已经存在的坏模式，遇到跨抽象层的复杂 Bug 时几乎没法自主解决。我们需要对后端或前端有深度理解的专家，作为代码质量的守门人，在 AI 输出的内容合并到主干之前做严格的审核，避免整个代码库随着 AI 的广泛使用慢慢烂掉。

Even engineering manager roles are getting a rewrite, the piece points out. All those AI tools that claim to evaluate engineer productivity and replace managers? They're almost all useless, because AI will never have the full context of a team's dynamics, unspoken constraints, and individual growth goals. What we will see a lot fewer of, though, are managers who don't write any code. With teams shrinking to 2-3 people, managers spend far less time on people management overhead, so they can jump back into coding work, using AI tools to contribute to implementation without pulling away from their core management responsibilities.

最让我好奇的是，这些变化其实还只是 AI 冲击团队结构的第一波浪潮。文章的分析甚至是建立在「AI 能力不会再大幅提升」的保守假设上的，如果未来产品和设计方向的 AI 工具也迎来了质的突破，我们的组织架构会不会还要经历更彻底的重构？说不定再过几年，我们回头看现在的团队配置，会觉得和当年的瀑布式开发流程一样过时。

来源：https://www.jampa.dev/p/the-rise-of-one-pizza-engineering